Geneva's 90% Stool Test Could Make Colonoscopies Optional

A Stool Sample That Rivals a Camera Inside Your Colon

Colorectal cancer is the second leading cause of cancer-related deaths worldwide, and its incidence among younger adults is climbing. The gold standard for detection remains colonoscopy — an invasive, preparation-heavy procedure that many patients delay or avoid entirely. Every non-invasive alternative developed so far has traded convenience for accuracy, leaving clinicians in a persistent bind: the best test is the one patients will actually take, but that test has never been good enough.

That calculus may be shifting. A team at the University of Geneva (UNIGE) has demonstrated that analyzing gut bacteria in stool samples at an unprecedented level of biological detail — the subspecies level — can identify 90% of colorectal cancer cases. That figure approaches colonoscopy's 94% detection rate and, according to the researchers, outperforms every non-invasive screening method currently available. The work, published in Cell Host & Microbe, introduces both a new diagnostic concept and a practical tool that could reshape how the world screens for one of its deadliest cancers.

The Resolution Problem: Why Species-Level Analysis Falls Short

The human gut contains trillions of bacteria spanning hundreds of species. Over the past decade, researchers have established that the composition of this microbial community — the microbiome — shifts in characteristic ways in patients with colorectal cancer. The idea of using stool-based microbiome analysis as a diagnostic tool is not new. What has been missing is sufficient resolution.

Most microbiome studies classify bacteria at the species level: Bacteroides fragilis, Fusobacterium nucleatum, and so on. But within a single species, different variants can behave in fundamentally different ways. Some strains of a given species may promote tumor development, while sibling strains of the same species are benign or even protective. When researchers lump all variants together under one species label, the cancer-associated signals get diluted by noise from their harmless relatives.

Going to the opposite extreme — strain-level analysis — introduces a different problem. Strains are so specific to individual hosts and environments that patterns found in one study population often fail to replicate in another. A strain identified as cancer-associated in a European cohort may not even appear in an Asian one, not because the biology differs but because strain-level classification is too granular to generalize.

The UNIGE team recognized this as fundamentally a resolution problem. The diagnostic signal exists in the microbiome, but you need to look at the right zoom level to see it consistently.

The Subspecies Solution: Building the HuMSub Catalogue

The study's lead author, Matija Trickovic, a PhD student in the laboratory of Professor Mirko Trajkovski at UNIGE's Department of Cell Physiology and Metabolism, built what the team describes as the first comprehensive catalogue of human gut microbiota subspecies — a resource they call HuMSub.

The approach defines operational subspecies units (OSUs) by clustering bacterial genomes based on their coding sequences in an unbiased, cohort-independent manner. This creates groupings that sit between the too-broad species level and the too-narrow strain level. The researchers also developed a computational method called panhashome, a sketching-based tool that enables rapid subspecies quantification and identifies the specific genes that drive functional differences within species, as described in the study abstract.

The key finding is that subspecies carry what the researchers call "implicit information" — disease associations that are invisible at the species level. The study identified subspecies associated with colorectal cancer whose sibling subspecies, and even their parent species, showed no such association. In other words, the cancer signal was hiding inside the species, visible only when you split the species into its constituent subspecies.

"The subspecies resolution is specific and can capture differences in how bacteria function and contribute to diseases including cancer, while remaining general enough to detect these changes among different groups of people, populations, or countries," Trajkovski explained in the UNIGE press release.

The Diagnostic Algorithm: From Catalogue to Cancer Detection

With the subspecies catalogue established, the team built a machine learning classifier trained on a meta-analysis of all available colorectal cancer patient datasets. The classifier takes subspecies-level microbiome data from a stool sample as input and outputs a cancer risk prediction.

The results were striking. "Our method detected 90% of cancer cases, a result very close to the 94% detection rate" of colonoscopy, Trickovic reported in ScienceDaily. Crucially, the subspecies-based machine learning model outperformed traditional species-level approaches applied to the same data — demonstrating that the improvement comes specifically from the higher-resolution analysis, not simply from better algorithms.

To appreciate what 90% sensitivity means in context, consider the current landscape of non-invasive colorectal cancer screening. The fecal immunochemical test (FIT), which detects blood in stool, is the most widely used non-invasive screening tool globally. Multi-target stool DNA tests add molecular markers to improve on FIT's performance. The Geneva team's press materials describe their method as outperforming all such alternatives, placing subspecies-level microbiome analysis at the top of the non-invasive hierarchy — closer to colonoscopy than any stool-based test has come before.

What This Test Could Look Like in Practice

If the approach translates to clinical use, the screening workflow would change significantly. Rather than choosing between colonoscopy (accurate but invasive) and existing stool tests (convenient but less sensitive), patients could begin with a subspecies-level microbiome analysis of a routine stool sample. Those who test negative would be spared the preparation, sedation, and procedural risks of colonoscopy. Those who test positive would proceed to colonoscopy for confirmation and, if necessary, polyp removal or biopsy.

This tiered model — microbiome screening as first gate, colonoscopy as follow-up — could substantially increase screening compliance. Colonoscopy adherence rates remain a persistent public health challenge. In many countries, participation in organized screening programs hovers well below recommended levels, partly because the procedure itself deters patients. A stool test that catches 90% of cancers could serve as the low-friction entry point that colonoscopy has never been.

The scalability argument is equally compelling. Colonoscopy requires trained endoscopists, specialized facilities, and sedation infrastructure. In resource-limited settings — and across much of the developing world — this infrastructure simply does not exist at the scale needed for population-level screening. A stool-based test that requires only sample collection and computational analysis could extend meaningful cancer screening to populations that currently have no access, as Bioengineer.org noted.

The Caveats: What 90% Does Not Mean

Before declaring colonoscopy obsolete, several important limitations deserve attention.

First, 90% sensitivity means 10% of cancers are missed. For a disease where early detection is the difference between a routine outpatient procedure and months of chemotherapy, a 10% miss rate is not trivial. Colonoscopy's 94% sensitivity is itself imperfect — no screening test catches everything — but the gap between 90% and 94% represents real patients with real cancers that would go undetected.

Second, the study's performance figures come from a meta-analysis of existing datasets, not from a prospective clinical trial. A meta-analysis can demonstrate that the signal exists and that the algorithm works on retrospective data, but it cannot tell us how the test performs in the messy reality of clinical practice — where sample handling varies, patient populations differ from study cohorts, and the prevalence of cancer in a screening population is much lower than in curated research datasets.

The researchers are aware of this gap. A clinical trial is being established in collaboration with the Geneva University Hospitals (HUG) to determine more precisely which cancer stages and which types of lesions the method can reliably detect, according to the UNIGE announcement. Until those results arrive, the 90% figure should be understood as a proof-of-concept benchmark, not a clinical performance guarantee.

Third, sensitivity is only half the equation. Specificity — the test's ability to correctly identify people who do not have cancer — is equally critical for a screening tool. A test with high sensitivity but poor specificity would generate enormous numbers of false positives, sending healthy patients to unnecessary colonoscopies and overwhelming the very system the test is meant to relieve. The available press materials and abstract do not prominently feature specificity figures, which will be essential to evaluate once the clinical trial data emerge.

Fourth, the test detects colorectal cancer, not precancerous polyps. Colonoscopy's greatest advantage is not just detection but prevention: during the procedure, clinicians can identify and remove polyps before they become cancerous. A microbiome-based stool test, however accurate for cancer detection, cannot replace this interventional capability. The question becomes whether the test can also detect advanced precancerous lesions — a question the upcoming clinical trial is designed to answer.

Beyond Colorectal Cancer: A Platform, Not Just a Test

Perhaps the most far-reaching implication of the UNIGE work is not the colorectal cancer test itself but the underlying platform. The HuMSub catalogue and panhashome methodology establish a general framework for analyzing the microbiome at subspecies resolution. Colorectal cancer was the validation case, but the same approach could, in principle, be applied to any condition where the gut microbiome carries diagnostic information.

Research has linked microbiome composition to inflammatory bowel disease, metabolic disorders, liver disease, neurological conditions, and even treatment response in cancer immunotherapy. In each of these areas, the same resolution problem applies: species-level analysis may be too blunt, strain-level too noisy, and subspecies-level analysis may unlock signals that have been invisible to both.

The pending patent application filed by Trickovic, Kieser, and Trajkovski — covering the use of subspecies in diagnostics and personalized medicine — signals that the team sees this as a platform technology with applications well beyond a single cancer type. If the subspecies framework proves as generalizable as the researchers suggest, a single stool sample analyzed with this methodology could eventually screen for multiple conditions simultaneously.

This is speculative, and the researchers themselves have not made such broad claims in their published work. But the architectural possibility is there: the HuMSub catalogue is not a colorectal cancer tool that happens to use subspecies data. It is a subspecies data infrastructure that happens to have been validated first on colorectal cancer.

The Competitive Landscape: Where Microbiome Diagnostics Stand

The Geneva team is not alone in pursuing microbiome-based cancer diagnostics. Several companies and research groups have explored gut microbiome analysis for colorectal cancer detection, though most have worked at the species level and reported lower sensitivity figures.

What distinguishes the UNIGE approach is the combination of three elements: the subspecies resolution (which improves signal quality), the comprehensive and cohort-independent catalogue (which enables generalization across populations), and the machine learning meta-analysis across all available CRC datasets (which maximizes statistical power). No previously published approach has combined all three.

The regulatory path for microbiome-based diagnostics remains largely uncharted. Unlike molecular biomarkers with well-established regulatory frameworks, microbiome-based tests involve complex, multi-analyte signatures that do not fit neatly into existing diagnostic categories. How regulators in the United States, Europe, and Asia choose to classify and evaluate these tests will significantly influence how quickly they reach patients.

Implications: What Changes If This Works

If the clinical trial confirms the meta-analysis results, the implications cascade across several dimensions.

For patients, the most immediate impact is accessibility. A non-invasive test that catches nine out of ten cancers from a stool sample lowers the barrier to screening dramatically. Patients who have avoided colonoscopy — whether due to discomfort, cost, lack of access, or simple procrastination — gain a meaningful alternative.

For healthcare systems, the economics shift. Colonoscopy is expensive: the procedure itself, the sedation, the facility time, the specialist's time, and the lost productivity for the patient who must take a day off work. If a significant fraction of screening colonoscopies could be replaced by an upstream microbiome test, the cost per cancer detected could fall substantially — freeing resources for patients who genuinely need the invasive procedure.

For the microbiome research field, the study validates a methodological principle. The demonstration that subspecies-level analysis unlocks diagnostic signals invisible at the species level provides a template that other researchers can apply to other diseases. If replicated across conditions, this could accelerate the translation of microbiome research from academic curiosity to clinical utility.

And for global health equity, the scalability of a stool-based test matters most. In countries where colonoscopy infrastructure is limited or nonexistent, this approach could bring colorectal cancer screening to populations that currently have none. The gap between what is technically possible in a well-equipped hospital and what is actually available to a patient in a rural clinic remains one of the defining challenges of global cancer control. A test that requires a stool sample and a computational pipeline — but not an endoscope — could begin to close it.

Key Takeaways

• A University of Geneva team created the first comprehensive subspecies-level catalogue of human gut bacteria (HuMSub) and used machine learning to detect colorectal cancer from stool samples with 90% sensitivity, approaching colonoscopy's 94% rate.
• The breakthrough lies in resolution: subspecies-level analysis captures cancer-associated microbial signals that are invisible at the species level, where most prior microbiome studies have operated.
• The method outperformed species-level machine learning approaches on the same data, confirming that the improvement comes from finer biological resolution, not just better algorithms.
• A clinical trial with Geneva University Hospitals is being established to validate performance in real-world screening and determine which cancer stages and precancerous lesions can be detected.
• The underlying platform — subspecies catalogue plus panhashome computational tool — could extend to diagnostics for other diseases linked to microbiome composition, potentially enabling multi-condition screening from a single stool sample.